"Copy It"

Late in 1961, a thin man with dark hair and deep-set eyes stepped off a transatlantic flight at Sheremetyevo with a small object hidden in his luggage. Boris Malin had spent a year as a Soviet exchange student in Pennsylvania, and he was bringing home a souvenir no customs officer could possibly have appreciated: a Texas Instruments SN-51 integrated circuit, one of the very first ICs ever sold in the United States. The chip was no larger than a thumbnail. It contained a handful of transistors fused onto a single sliver of silicon. In the West it cost a few dollars. In Moscow it was about to become the most studied object in Soviet electronics.

A few weeks later, Malin sat across a desk from Shokin. Shokin took the chip, placed it under a microscope, and looked. According to engineers who were there and to histories drawn from Soviet sources, what he said next set the course of Soviet microelectronics for thirty years. Copy it, he told Malin. One-for-one. No deviations. You have three months.

Malin returned to NII-35, the Scientific Research Institute later known as Pulsar, where Department 5 reported to him directly. The department had roughly two hundred and fifty engineers and an experimental workshop. They began the work that would define Soviet semiconductors for the rest of the Cold War: depressurizing the package of a Western chip, lifting off its lid, photographing the planar pattern of doped regions and metal traces under a microscope at successive depths, and reconstructing the working drawings by hand. The technique was called decapping, a word that captures something of its small violence. By the late 1960s, after pilot runs at NII-35 between 1962 and 1965 and a transition to mass production at Fryazino outside Moscow in 1969, the SN-51 had become a Soviet part. By 1970, the West had moved on to TTL logic that made the SN-51 obsolete. Soviet factories kept making it anyway, in volume, until the country dissolved in 1991.

This was the pattern the entire industry would follow. It had a name in internal Ministry of Electronics Industry directives, captured later by historians as the “copy it” doctrine. It had political cover at the highest level. And it produced, with remarkable consistency, last year’s American chip, in last year’s American volumes, with last year’s American architecture, just as the Americans were rolling out the next.

Shokin was not a hack. Born in 1909, he had spent his career inside the Soviet electronics apparatus, first running radio plants during the war, then climbing the ministerial ladder. In 1961 he chaired the State Committee for Electronic Technology; by 1965 he would be the founding minister of a new full ministry, the Ministry of the Electronics Industry, which he would run for two decades. He was, in the language of Soviet bureaucracy, a builder. But the same impulse that had produced Zelenograd, the conviction that microelectronics could be willed into existence by central decision, also produced the copy-it order. If a Western chip already worked, there was no need to wait for Soviet engineers to design something equivalent. The shortcut was obvious. The shortcut was also, in microelectronics, a trap.

The doctrine hardened in the late 1960s and was made explicit at the level of mainframes a few years later. In 1966, Soviet economic planners began debating how to standardize the country’s chaotic computing sector, in which dozens of incompatible architectures, the BESM line, the Minsk line, the Ural line, jockeyed for resources. Some of the Soviet Union’s most distinguished computer scientists, including the BESM designer Sergei Lebedev, argued that the country should pick its best indigenous machine and double down. They lost. In 1968, Brezhnev’s Politburo, with the Ministry of Radio Industry leading the technical case, decided instead that the Eastern Bloc would clone IBM’s System/360. The cloned mainframe was christened the ES EVM, the “Unified System of Electronic Computing Machines,” and known throughout the Comecon countries as Ryad, “Series.” Bulgaria, Czechoslovakia, East Germany, Hungary, and Poland were strong-armed into participating. Indigenous computer lines were starved of funding and quietly shut down.

In 1972, the year the first Ryad machines shipped, Brezhnev gave the doctrine its blunt one-liner. The Soviet Union, he told officials, would have to string along with the capitalists for a while. We need their credits, he said, their agriculture, and their technology. The remark, widely reported in Western intelligence summaries afterward, treated Western technology as a resource to be extracted, like wheat or hard currency, rather than as the output of a system that had to be understood and reproduced. You could buy grain and ship it home. You could not, it would turn out, buy Moore’s Law.

What followed in microelectronics was a meticulous, decades-long campaign of cloning that produced a parallel parts catalog mirroring the West with a delay of three to five years. Soviet engineers cloned the Intel 8080. They cloned the Intel 8086. They cloned the Zilog Z80. They cloned the DEC LSI-11. They cloned the AMD 2901 bit-slice. They cloned the Motorola 6800 in Bulgaria and the MOS 6502 in Bulgaria, the Intel 8048 in Russia, and a long list of supporting parts, memories, peripheral controllers, analog interface chips, that filled out the catalog. The Soviet K580 family began life as the K580IK80, a clone of the 8080, with the project initiated in August 1976 at the Kiev Institute of Microdevices, two years after Intel released the original. Prototypes followed in 1977 and 1978; renamed KR580VM80A under the 1980 Soviet integrated-circuit naming standard, it was eventually produced at Kvazar in Kiev, Angstrem in Zelenograd, Rodon in Ivano-Frankivsk, Dnepr in Kherson, Electronpribor in Fryazino, and Kvantor in Zbarazh. The K1810VM86, a clone of Intel’s 8086, was developed between 1982 and 1985 and became the workhorse processor of the late-Soviet personal computer scene, the only x86-compatible chip available to Eastern Bloc engineers who wanted to run MS-DOS. The 1801 series, beginning with the K1801VM1 in 1982, replicated the instruction set of the DEC LSI-11 and made Soviet PDP-11 clones possible. By the time the Soviet Z80 clone shipped in 1990, designed at the Scientific Research Institute of Precise Technology and the Angstrem plant in Zelenograd, the West was already three generations beyond it.

The actual engineering work was both more sophisticated and more thankless than the word “copying” suggests. To turn a Western chip into a Soviet chip, a team had to identify the package, dissolve or grind off its lid without damaging the die, photograph the metal layers, etch one layer at a time to expose the polysilicon and diffusion patterns underneath, photograph again, and assemble the resulting stack into a topological model of the circuit. They then had to recover the schematic from the topology, a non-trivial task, because a planar transistor layout does not announce its function. A reverse-engineered NAND gate looks more or less like a reverse-engineered NOR gate to a microscope. Engineers had to prove the equivalence by testing.

By the late 1980s, Western chip designers had begun to anticipate decapping and to fight back. When the Angstrem team set out to clone the Zilog Z80 under chief designer Yuri Otrokhov, who had served as a tank driver in his youth and who named the project T-34 after his old tank, the engineers found that Zilog had laid traps. A circuit cell that looked under the microscope like a three-input NAND gate behaved electrically as a two-input NAND, the third input being a decoy. The deceptions were structural. You could only catch them by using probe analyzers on the working silicon and comparing measured logic to apparent layout. Otrokhov’s team mapped the traps, neutralized them, redesigned the layout in a tighter two-micron NMOS process where Zilog had used four-micron, and turned out a working clone in nine months across four design iterations. The Soviet T34VM1, later renamed KR1858VM1, was an impressive piece of reverse engineering. It also borrowed dies and photomasks from the East German U880, itself a Z80 clone produced by MME, suggesting that the workload had been quietly distributed across the bloc. By the time the team was done, in 1990, Zilog’s actual customers in the West had been buying CMOS Z80s for a decade and the chip was a legacy part. The clock did not stop because the Soviet team was good. It only stopped because the original had aged out of relevance.

That was the structural problem. In a discrete-engineering field, the kind that produced Soviet rockets and Soviet bombs, copying scaled. The first atomic bomb tested at Semipalatinsk in 1949 was a near-replica of Fat Man, assembled with intelligence supplied by the Manhattan Project spies, and it worked. The hydrogen bomb followed indigenously a few years later. Microelectronics was not like that. It ran on the cadence Gordon Moore had identified in 1965: a doubling of components per chip roughly every eighteen to twenty-four months, sustained over decades, supported by an ecosystem of process improvements, lithography refinements, materials advances, and design tools that fed each other in a tight loop. To clone an Intel chip, a Soviet team needed perhaps two to four years from the moment a working sample arrived in Moscow. Intel, in those same two to four years, was already shipping its successor. The lag was not a fixed offset. It was a treadmill. The faster the cloning got, the faster the original ran.

This is the geometry that made the copy-it doctrine an engine of permanent inferiority. A 1986 declassified CIA assessment titled Soviet Microelectronics: Impact of Western Technology Acquisition concluded that the Soviet industry trailed the West by roughly five to seven years across most product categories, and that the lag was widening, not narrowing, despite a vast and well-funded acquisition program. The Soviet Union had, by then, organized something close to a state-scale technology-extraction industry. KGB Directorate T, the scientific and technical intelligence arm, ran an operation called Line X across more than a hundred Western embassies; its officers, posing as diplomats, recruited engineers and bought blueprints. The Ministry of Electronics Industry maintained shopping lists of specific tools and chips it wanted. Front companies in neutral countries bought equipment that violated the COCOM, the Coordinating Committee for Multilateral Export Controls, restrictions and routed it home through Sweden, Austria, and Switzerland. When the French intelligence service DST received the so-called Farewell Dossier from KGB Colonel Vladimir Vetrov in 1981 and 1982, it counted nearly four thousand internal Soviet documents listing the specific Western technologies the bloc was hunting and the means by which it had acquired them. Semiconductors and computers headed the list. The program was enormous, and still could not close the gap.

The contrast with Japan, working the same problem in the same years, made the difference inescapable. Japanese engineers also reverse-engineered Western chips and Western tools throughout the 1960s and 1970s. NEC, Hitachi, Toshiba, Mitsubishi, and Fujitsu all had teams that took apart American DRAMs and American steppers and learned how they worked. But the Japanese cloning was the entry point, not the destination. When the consortium known as the VLSI Project ran from 1976 to 1979 under MITI, the major firms collaborating on next-generation lithography and process development, the explicit goal was to use the lessons of reverse engineering to leapfrog the originals. By the 1982 transition to 64-kilobit DRAMs, Japanese firms held seventy percent of the world market. By the 256-kilobit generation in 1984, the share was ninety percent. By the megabit generation in 1988, the share was still ninety percent and the Americans were panicking. Japan had used copying to learn, and then it had run.

Soviet engineers were not less talented than Japanese engineers. The Zelenograd labs included some of the finest semiconductor physicists of the postwar era, and individual Soviet design achievements, including indigenous bipolar logic families and radiation-hardened parts for spacecraft, showed real originality. The difference was systemic. Japan had, however imperfectly, a market mechanism that punished factories making yesterday’s chip and rewarded factories making tomorrow’s. It had thousands of customers shouting for incremental improvement. It had supplier relationships that ratcheted process technology forward. The Soviet electronics ministry had a five-year plan that specified a quota of K580 chips and a quota of K1801 chips and rewarded the directors who hit those quotas. There was no internal customer who could refuse a chip for being obsolete. The plan said make this chip. The chip got made. That the West had moved on was, from the standpoint of the production target, a problem someone else would have to solve next year, or the year after, by writing a new plan that said clone the next chip.

The historian Slava Gerovitch, writing about Soviet cybernetics, captured something of how the language of Soviet planning made this trap invisible to the people inside it. Soviet officials had spent decades operating in what Gerovitch called Newspeak, a vocabulary in which slogans, plan fulfillments, and ideological formulations did the work that information was supposed to do. The cybernetics movement of the 1950s and 1960s had tried to introduce a new vocabulary, Cyberspeak, that promised objectivity and feedback. But by the late 1960s the two had merged into what he called CyberNewspeak, in which the discourse of computers and feedback was used to dress up the same old planning targets. Within that discourse, a chip that was three years behind the West but met its plan target was a success. A chip that was at the frontier of world technology but had failed to ship enough units was a failure. The frontier was, definitionally, somewhere else.

There was also the awkward fact that the Americans had figured out roughly what was happening and had begun to push back. After Vetrov’s documents reached Washington through French channels, the Reagan administration approved an operation, run through the National Security Council under Gus Weiss, to feed the Line X collection network deliberately doctored technology, including, by some accounts, sabotaged software for a Soviet gas pipeline and faulty chips routed through the same channels the KGB was using. The CIA, according to declassified documents and to later memoirs, did not need to do much of the sabotage actively. The act of cloning a chip that had already been altered to fail under Soviet manufacturing tolerances was enough. By the mid-1980s Western intelligence officers spoke quietly of a Soviet electronics sector that was, in effect, building Moscow’s hardware on Washington’s terms.

Barr and Sarant watched the doctrine harden from inside. The two engineers had originally pitched Zelenograd as a place where Soviet talent would invent. By the late 1960s, with Shokin’s directives in full force, what they had actually built was a place where Soviet talent reproduced.

The Soviet engineers themselves understood what was happening. Soviet engineering memoirs from the period, collected and translated in the years after 1991, return again and again to the same observation: the team finished the project, the chip worked, and by the time it shipped the world had already moved on. They were running a relay race in which the baton kept being passed forward by the runner ahead. There was no way, by running harder, to catch up. To catch up, they would have had to be the runner ahead.

Late in his life, Boris Malin gave interviews in which he described the moment in 1961 when he had carried the Texas Instruments chip back from Pennsylvania. It had felt, he said, like a beginning. It had been, in fact, the beginning. The order Shokin gave him that afternoon, copy it, was the order an entire generation of Soviet engineers would receive in some form, and the order they would obey faithfully, with skill, for the next thirty years. They would copy almost everything the Americans built, with growing virtuosity, decapping more cleverly, neutralizing copy-protection traps, shrinking process nodes when they could. And they would never, not once, ship a chip that was at the frontier. The frontier was not a thing you could put under a microscope. It was a velocity. You could not espionage your way to a velocity. You had to invent your way there, and the Soviet system, for reasons that ran deeper than any one minister’s directive, was not built to invent.

In Tokyo, in those same years, the lesson was being learned in the opposite direction.